Factory IoT Cost 2026 | The Five Layers That Split Quotations

Ask three vendors to quote the same factory IoT project and you can end up holding proposals as far apart as the 3,500,000 yen and 15,000,000 yen that mark the low and high ends of the range published in Japan, with sources given later in this article. The sensor count is roughly identical in both. The totals differ by an order of magnitude. That gap is not a pricing problem. It comes from the fact that each vendor has drawn the boundary of the quotation in a different place, and the parts left outside that boundary are never written down. This article breaks factory IoT cost into five layers, shows which layer sets the order of magnitude and which layer quietly generates a second round of spending, and works the whole thing through with a cost build-up for 30 machines at a plant in Thailand.

The real reason factory IoT quotations split by an order of magnitude

Most articles written about factory IoT cost end at “it depends on how many machines you have” and “it depends on your requirements”. Neither statement is false, but neither is any use at all to the person holding three quotations and trying to compare them. Quotations for the same machine count still split.

The reason is simple. A factory IoT project bundles five completely different kinds of work into a single quotation, and each vendor includes a different subset of those five in scope. A cheap quotation is not the result of a vendor cutting corners. It is the result of a vendor leaving out layers. And those layers do not disappear. After the contract is signed, somebody performs that work: your own production engineering team, or the vendor under a change order you pay for, or nobody at all, in which case the project stops at a dashboard and never becomes anything more.

So the first thing to do when reading quotations is not to line up the totals. It is to go through them one vendor at a time and confirm which of the five layers each one contains. Comparing the money on quotations whose layers do not match is not a comparison at all.

There is a second thing that distorts the cost conversation. Factory IoT cost tends to get discussed as a price per machine, and that denominator does not reflect reality. Take three machines. One exposes readable tags over Ethernet. One offers nothing but a dry contact from its signal tower light, the stacked red, yellow and green lamp also known as an andon or stack light. One has an analogue gauge with a needle and nothing else. The work required on each is completely different. A unit price using machine count as the denominator averages that difference away until it vanishes. Later in this article we set out the three variables that belong in the denominator instead.

Breaking factory IoT implementation cost into five layers

Factory IoT cost breaks down into the following five layers. The order of the layers is also a dependency order, in the sense that until the earlier layer is settled the later ones cannot be.

The sections below go through each layer and what moves the money inside it.

Layer 1 – the signal source sets the order of magnitude

Layer 1 is the form in which each target machine currently emits a signal. Nothing you negotiate or engineer changes this. It was fixed on the day the machine was purchased. In practice it falls into four patterns.

From the top of that table to the bottom, the work per machine varies by a factor of several times up to almost ten. What sets the order of magnitude of the total is how many of your 30 target machines sit in each pattern. It is not the machine count.

The one thing worth doing yourself before you request a single quotation is to count your machines against these four categories. Once you know the breakdown, you can read the assumptions sitting behind a vendor’s quotation. Request quotations without that breakdown and the vendor has no choice but to assume whatever suits them, and that assumption never appears in the document. Concrete methods for getting a signal out of machines you already own are covered in our article on IoT retrofit for legacy equipment.

Layer 2 – connection and installation work is what creates the spread

Deciding what to capture in Layer 1 does not settle how you physically extract it. That depends on the conditions at your site. Layer 2 is the installation layer, and it is the layer that creates the spread in the total.

Five things drive that spread.

• Whether control panel work is possible, and how it has to be staged. Can hands go into a panel while the machine is live, or does the circuit have to be shut down? At a site where the only opportunity to take a shutdown is the year-end holiday, installation scheduling becomes the critical path for the entire project.
• The number of shutdown windows required. How many machines you can cover in one shutdown determines the number of man-days you need. If the target machines are spread across separate buildings, the electrical distribution differs by building and the number of windows goes up.
• Area classification. If any target machine sits in a hazardous area, a cleanroom or a temperature-controlled room, both the equipment selection and the work procedure change. Hazardous areas restrict the equipment options sharply, and unit prices jump.
• Warranty on existing machines and contracts with the machine builder. If a third party works inside the control panel of a machine still under a maintenance contract, the warranty can be voided. Some cases require the machine builder to attend, and that attendance fee is normally not included in a factory IoT quotation at all.
• Cable routing. No spare capacity in the existing cable tray, no access above the ceiling, a floor pit already full. None of this is knowable without walking the site.

Layer 2 cannot be estimated without a site survey. Put the other way round, any quotation produced without a site survey necessarily contains standard assumptions for Layer 2. When those assumptions do not match your site, the difference arrives as a change order after signature. Some vendors run a paid site survey first and credit the survey fee against the main contract. That arrangement is what allows the vendor to put people on your site in the first place, and for the buyer it is the healthier structure.

Layer 3 – the collection platform is easy to compare but small in money

Layer 3 is the mechanism that gathers and stores the signals you have captured. It covers gateways, the network, the collection server or cloud service, time synchronisation, and the buffer that holds data while communications are down.

Vendor specifications in this layer are similar to one another, which makes quotations easy to line up side by side. And because they are easy to compare, this is where most of the discussion time goes. It is not unusual to see meeting after meeting spent on gateway model numbers and on cloud versus on-premise, for a layer that is not a large share of the total.

What actually needs checking in Layer 3 is not the model number. It is these three points.

A quotation that says nothing about these three can look cheap in Layer 3, for the simple reason that a configuration with no buffer is cheaper to build.

Layer 4 – meaning is what the second round of cost really is

Layer 4 is the layer that gives meaning to the data you have collected. Specifically it covers the following design work.

• Tag design. The mapping table that says which signal represents which state on which machine. PLC tag names vary by machine builder and by the year the machine was installed, and in raw form they cannot be lined up next to each other.
• Equipment master. Equipment codes, lines, processes, and the reference values for capacity. Whether you adopt the same coding scheme as your existing production management system determines how much integration effort you face later.
• Units and reference values. Whether one production count means one piece or one lot. Where the reference cycle time comes from.
• Shift calendar. Two shifts or three, how breaks and lunch are handled, how Thai public holidays and company holidays get onto the calendar. This is where the denominator of your utilisation figure is decided.
• Downtime reason codes. The scheme that attaches a reason to the fact that something stopped. Build it at a granularity the shop floor will not actually use and the entry rate collapses.

You can skip Layer 4 and the system will still run. That is exactly the problem. A factory IoT deployment with Layer 4 omitted gives you a dashboard that shows whether the status lamp is red or green, and it cannot answer why the machine stopped. Roughly six months in, the shop floor says that this is not usable for improvement work, and the second budget request begins.

The second round is not necessarily smaller than the first. Retrofitting meaning into a system that is already live takes more effort than designing it on a clean sheet. Change the coding scheme in the equipment master and all the historical data has to be remapped. Introduce downtime reason codes after the fact and you have to build the entry mechanism and run shop floor training from zero.

When comparing factory IoT cost, whether Layer 4 is in the quotation is the second most important thing to check, after Layer 1. Quotations that exclude it always look cheaper.

Layer 5 – the user-facing side is where requirements swell

Layer 5 is the part people actually touch. Dashboards, daily and monthly report formats, alerts when something goes wrong, and integration with the production management system or ERP.

The defining characteristic of this layer is that requirements grow after the fact. What starts as “we just want to see utilisation” becomes, once it is running, “can we see it by product”, “can we see it by team”, “can we compare against last month”. That growth is evidence the system is being used, so it is not a bad thing in itself. But putting all of it into the initial quotation inflates the total and the capital request will not clear internal approval.

The practical answer is to quote Layer 5 in two stages from the outset. Stage one is a single screen that gets looked at daily, and nothing else. Stage two is added once operations are running, based on which screens people genuinely use. This split has the same structure as the small start approach to system implementation, and if there is one layer where a small start works best, Layer 5 is it.

Layers 1 through 4, by contrast, are poor candidates for a small start. Signal source classification, installation staging and tag design all generate rework when you widen the scope later. Narrowing the number of target machines is fine. Narrowing the number of layers is dangerous.

Comparing on price per machine will always mislead – the denominator is not machine count

The most commonly used metric in any factory IoT cost discussion is price per machine, and it is also the most misleading. The five-layer model explains exactly why. Layers 1 and 2 are determined by the condition of each machine, and Layer 4 is determined not by how many machines you have but by how many distinct kinds of signal you have. Put machine count in the denominator and the information from all three of those layers is averaged out of existence.

What belongs in the denominator is a set of three variables.

Number of signal points

How many points you capture from one machine. Whether it is a single point saying running or stopped, three points once you add production count and reject count, or ten points including temperature, pressure and current, changes both the hardware point count in Layer 1 and the wiring effort in Layer 2.

In practice the point count grows in the following order. Wanting everything down to the bottom row from day one is a natural human impulse, but point count feeds straight into the total.

Sampling interval

How often you take a reading. A one-minute interval and a one-second interval differ in traffic volume, storage capacity, and how many machines a single gateway can carry. Move to one second and the number of machines per gateway falls, which increases the hardware count in Layer 3.

Work the interval backwards from the purpose. If all you need is a daily utilisation figure, one minute is enough. If you want to catch minor stoppages lasting tens of seconds, you need an interval between one and five seconds. The reasoning behind that judgement is covered in our article on minor stoppages and OEE. Deciding to “capture it finely just in case” quietly pushes up Layer 3 cost.

Tolerance for missing data

How much data loss you can accept. This drives real money, yet it is almost never written down as a requirement.

• Missing data is acceptable. The use case is a daily trend and nothing more. A cheap configuration with no buffer is sufficient.
• Missing data is acceptable, but the fact that it went missing must be recorded. This is needed so the gap can be excluded from the utilisation calculation. It requires log design at the gateway.
• Missing data is not acceptable. This applies when the data feeds traceability or quality records. You need buffering, retransmission or redundancy, and Layer 3 cost changes accordingly.

Issue an RFP without settling these three variables and every vendor assumes something different. A quotation built on 2 points, one-minute interval and tolerated data loss comes back alongside one built on 6 points, one-second interval and zero tolerance, and the cheaper one gets selected. Then, once requirements are pinned down after signature, it turns out the first set of assumptions cannot satisfy them. This is the single most frequent accident in factory IoT procurement.

Cost build-up for 30 machines at a Thai plant – three scenarios

From here we work through a concrete build-up modelled on a Japanese-owned plant in Thailand. One warning first. The figures below are a calculation example built on assumptions stated in this article. They are not market rates. Real quotations vary widely with site conditions and specification. Do not copy these numbers straight into a budget.

Assumptions and the unit rates used

The assumptions are as follows.

• The site is a Japanese-owned plant in Thailand with 30 target machines. Electricity metering is out of scope for this article.
• Installation work is inside an existing building. Hazardous areas and cleanrooms are excluded.
• The basis for labour cost is the Bangkok minimum wage of 400 baht per day. That rate took effect on 1 July 2025, replacing 372 baht, and provinces including Chonburi and Rayong moved to 400 baht from January 2025. For 2026 the rates remain unchanged, in a band from 337 to 400 baht.
• The engineering day rate is set separately from the minimum wage, as a blended figure covering local vendor engineers and in-house engineers, assumed by this article.

The unit rates used are below. All of them are assumptions made by this article.

Scenario A – capture the signal tower contact only

Take one dry contact from the signal tower light on each of the 30 machines and record nothing but running and stopped. This is the cheapest way to start.

Annual maintenance at 15 percent of 920,000 is 138,000 baht, or 414,000 baht over three years. Initial cost plus three years of maintenance comes to 1,334,000 baht.

The structural problem with this scenario sits in Layer 4. A dry contact from a signal tower light tells you that the lamp was red. Whether red meant a tool change, a material shortage or a machine fault is not contained in the contact. You can therefore produce a utilisation figure, but you cannot identify what to improve. Within six months to a year, additional work to attach reasons to stoppages will be on the agenda.

Scenario B – direct PLC connection

The assumption here is that 15 of the 30 machines have a PLC with Ethernet, and the remaining 15 are serial or have undocumented tags. In Japanese-owned plants in Thailand, this mixed state is by far the most common outcome of repeated expansion over the years.

Annual maintenance is 215,100 baht, or 645,300 baht over three years. Initial cost plus three years of maintenance comes to 2,079,300 baht.

What moves the total in this scenario is Layer 2, not Layer 1. Control panel work and shutdown scheduling for the 15 non-capable machines account for 304,000 baht, and that portion moves up or down with site conditions. At a site where a shutdown can only be taken twice a year, the number of windows rises and the man-days accumulate. Conversely, if the work can be aligned with a scheduled equipment upgrade, this layer compresses considerably.

The reason Layer 4 carries 20 man-days for tag design is that PLC tag names differ machine by machine. The tag representing “in operation” is defined under a different name on machines from a different builder. Mapping those onto a single unified state definition in the equipment master is, in substance, the core of this scenario.

Scenario C – full retrofit across all machines

Take nothing from the PLCs. Instead, retrofit external sensors onto all 30 machines. Because you never touch the machine builder’s control system, the existing warranty question is avoided entirely and the installation is unaffected by machine upgrades.

Annual maintenance is 277,200 baht, or 831,600 baht over three years. Initial cost plus three years of maintenance comes to 2,679,600 baht.

Layer 4 is largest in this scenario because logic has to be designed to infer machine state from raw sensor values. At what current draw does the machine count as stopped? How short a fluctuation should be ignored? Threshold design is tedious work, but the flip side is that you have no choice but to do it from the beginning. Downtime reason codes get built into the design at the same stage.

Three-year total within the scope of the quotation

Here are the three scenarios lined up on initial cost and three years of maintenance.

This table is a calculation example built on assumptions made by this article. It is not a market rate.

The ranking is A, then B, then C. Judged only on what appears inside the quotation, Scenario A wins, and in practice that is the decision many plants make. The next section shows what overturns that ranking.

Re-rank on the total cost of reaching a usable state and the order changes

The table above leaves out two costs. The second round of Layer 4 spending, and the human effort required to operate the system. Neither appears on a quotation. Both get paid.

Here is the second round of Layer 4 cost by scenario.

Operating effort differs too. Scenario A ends up with a configuration where a person enters the downtime reason by hand, so shop floor follow-up is needed every year to keep the entry rate up. Scenario C has automated determination, so this effort is almost nil.

Now we line up the three-year total cost of ownership including those items, alongside the number of months it takes to reach a state where downtime logs with reasons attached are in daily use. The months to reach that state are a figure assumed by this article, covering design, installation, commissioning and shop floor adoption across Layers 1 through 5.

This table is also a build-up by this article, not a market rate.

The ranking flips on the last row. On three-year total cost of ownership the order is A, B, C. Ranked on cost per usable month it becomes B, C, A, and the scenario that started cheapest ends up the most expensive.

The reason is not the size of the spend. It is time to reach usefulness. Scenario A discovers ten months after go-live that all it knows is the fact that something was stopped, and then spends another ten months designing reason codes and training the shop floor. For those 20 months the 920,000 baht already committed is not being used for improvement. Scenario C costs twice as much up front but enters a usable state in month eight.

Three practical conclusions follow.

• Select on initial cost alone and you will select the configuration that defers Layer 4. Layer 4 can be deferred because it causes no trouble at the quotation stage, but it becomes unavoidable the moment operations begin.
• Define “usable state” numerically before you compare anything. This article uses “downtime logs with reasons attached, in daily use”, but the definition can legitimately differ site by site. Without a definition, this comparison cannot be performed at all.
• Months to reach that state is something a vendor can answer if asked. Require every RFP response to state the expected number of months to enter daily operation, with Layer 4 included.

Where to put the return – measure it as reduced downtime

Having discussed cost, the return needs discussing too. The first justification usually offered for factory IoT is energy reduction, but a configuration built for machine status monitoring is usually not measuring electricity at all, so that justification does not hold. For status monitoring, the honest place to put the return is reduced downtime.

The skeleton of the calculation is simple.

Annual benefit = number of target machines x monthly downtime reduction per machine x 12 x value of one hour of downtime

The problem is the third variable. Depending on what you put in for the value of an hour of downtime, the answer moves by a factor of 40. Any document that leaves that number vague and then states a two-year payback offers nothing you can verify.

This article states the floor explicitly. The Bangkok minimum wage of 400 baht per day, divided by an eight-hour day, is 50 baht per hour. If one operator stands idle for the hour the machine is down, the wage lost is 50 baht per hour. That is the floor.

But do not use that 50 baht per hour as your basis for return. The real loss is not the wage. It is dominated by the gross margin on the product that hour could have produced, in other words the foregone profit. The more expensive the machine, and the fuller the order book, the wider that gap becomes. Fifty baht per hour is a floor meaning “at minimum this much is being lost”. It is not a number to base an investment decision on.

Here is the sensitivity across three assumed values.

Figures are in baht per year. This table is also a calculation example by this article, not an observed result.

Read it like this. The three-year total cost of ownership for Scenario B above was 2,451,300 baht. Value one hour of downtime at 500 baht and achieve five hours of reduction per machine per month, and the annual benefit is 900,000 baht, paying back in roughly 2.7 years. Achieve exactly the same reduction but value the hour at the wage-equivalent 50 baht and the annual benefit is only 90,000 baht, which works out at 27 years.

In other words, a factory IoT investment decision turns less on the precision of the cost estimate than on whether the value assigned to one hour of downtime is defensible. That number should already exist inside your finance and production control departments. Derive your own figure from standard cost, from the machine hour rate, or from the current order backlog. Use a generic figure from a vendor’s brochure and the payback calculation loses its basis at that moment.

One more caution, this time on the reduction side. Five hours per machine per month does not materialise simply because you installed factory IoT. Once the data exists, somebody has to look at the breakdown of stoppages and act on the largest causes. Data is a necessary condition for reduction, not a sufficient one, and that obvious point needs stating as a premise of any payback calculation. For how this has actually played out at sites in Thailand, see our article on manufacturing IoT case studies in Thailand.

Price ranges published in Japan, with sources

Everything above is this article’s own build-up for a site in Thailand. For reference, here is a summary of the cost ranges published in Japan. The currency and the assumptions differ from our build-up, so we neither add them together nor convert between them. Read them as separate information rather than as two halves of one picture.

The article by GXO, source 1, gives the following breakdown for the development cost of an IoT data collection and visualisation platform.

The same article also gives indicative implementation durations. Three to four months for a small deployment of 5 machines, six to eight months for a mid-sized deployment of 3 lines and 30 machines, and eight to twelve months for a company-wide rollout. Set against the 8 to 20 months this article assumed for reaching a usable state, the six to eight months quoted for a mid-sized 30-machine deployment is best read as a build period that does not include Layer 4 meaning or shop floor adoption.

The article by Three-up Technology, source 2, states the following on how cost is determined. A site survey runs to 150,000 yen for two days plus travel expenses, and is deducted from the implementation cost if the main contract is awarded. Installation work on a standard single machine takes about one day. And on the factors that decide cost, the article states explicitly that if the PLC supports Ethernet only an edge computer needs adding, whereas if it does not, electrical work is added on top.

That last statement is Layers 1 and 2 of our five-layer model, precisely. The same structure surfaces in how a Japanese domestic provider explains its own pricing. That article also states that there is no flat list price for factory IoT, and that cost is determined almost entirely by the condition of the machines.

What is worth taking from these two sources is not the ranges themselves but the explanation of why a range exists. The gap of more than four times between 3,500,000 yen and 15,000,000 yen is more likely a difference in which layers are included than a difference in pricing. If you use Japanese domestic figures as a reference point for your own budget expectations, you still have to confirm whether Layer 4 is inside that number, or it is not a comparison.

What separates a PoC that stalls from one that does not

Any conversation about factory IoT cost eventually reaches the proof of concept. “Let us start small and try it” is a sound proposal, but it is not unusual for a PoC to end at visualisation and for years to pass without budget for a production rollout.

The article by Next Vision, source 5, gives five structural reasons why factory DX stalls at PoC: vague success criteria, no scale-cost estimate, dependence on individuals, divergence from the existing systems, and a shop floor that was never made a participant. All five map onto the five-layer model in cost terms.

The decisive thing in PoC design is not narrowing the number of machines. It is running all five layers end to end. Three machines is fine, as long as you go from Layer 1 through Layer 5 and all the way to downtime reason codes. Do that and you come out holding measured effort figures for the production rollout.

The worst possible PoC, conversely, is one that skips Layer 4 and builds only Layers 3 and 5. The dashboard appears quickly so the PoC is well received, and then, at the point of quoting the production rollout, Layer 4 effort shows its face for the first time. At that moment the PoC cost is useless as a basis for the rollout quotation, and from the executive floor it looks like an unexplained situation where the PoC was cheap and production is expensive.

Three criteria separate a successful PoC from a failed one. Did downtime reason entry run as routine daily work on the shop floor for at least a month? Are the equipment master and tag design in a form that can be carried straight into the production rollout? Can per-machine effort for the rollout be estimated from the measured results? A PoC that cannot answer all three has not produced any material for the rollout decision.

Writing an RFP that makes quotations comparable

Everything above, translated into RFP form, becomes the following table. Write these items and multiple vendors’ quotations land on the same footing.

Of these, the two that are hard to fill in yourself are signal source classification and available shutdown dates. Signal source classification means reconciling the equipment register against the physical control panels, and at 30 machines it takes several days. Grudge those few days before issuing the RFP and you will spend weeks afterwards trying to compare quotations. Some vendors run a paid survey up front and credit the fee against the main contract. If you carve the survey out first, the survey result itself becomes an asset you own, and you can reuse it when requesting quotations from other vendors.

Thailand-specific considerations – installation, labour cost and BOI

Three assumptions differ between a site in Thailand and head office in Japan when considering factory IoT.

Installation staging and arranging people

Layer 2 installation will be contracted to local providers. What differs from Japan is that electrical work, network work and the IT-side work are frequently handled by three separate companies. Unless all three schedules align, the work planned for a single shutdown window cannot be completed. If project management effort is not written into the quotation, that coordination gets absorbed as overtime worked by the Japanese expatriate staff.

Also, where the machine builder’s local distributor is structured differently from its Japanese parent, it can take time to get an answer on whether attendance at the control panel is permitted. Confirming existing warranty conditions is something to begin before the RFP goes out, not after.

Labour cost and hourly rates

This article uses the Bangkok minimum wage of 400 baht per day as the basis for labour cost. That rate took effect on 1 July 2025, replacing 372 baht. Certain provinces, including Chonburi and Rayong, moved to 400 baht from January 2025. For 2026 the rates remain unchanged in a band from 337 to 400 baht, with no increase applied.

The engineering day rate sits at a different level from the minimum wage. This article assumes 8,000 baht per man-day, a figure blending vendor engineers and in-house engineers, and actual rates vary with the vendor and the nature of the work. If head office in Japan builds an estimate using Japanese domestic day rates, it will substantially overstate Layer 2 cost for a Thai site. Conversely, cheaper labour in Thailand does not make Layer 4 design effort cheaper. Layer 4 is design work spanning Japanese, Thai and English, and if anything the language dimension adds effort.

BOI investment incentives

The Thailand Board of Investment announced a new package of investment incentives on 15 January 2026. Within the framework for promoting technology upgrades, an additional three years of corporate income tax exemption and an import duty exemption on machinery are set out, and the eligible technologies explicitly include predictive maintenance systems and IoT sensor platforms.

A factory IoT investment may fall within that framework, but this article does not state deduction rates or specific monetary effects. Eligibility conditions, the scope of qualifying machinery and application deadlines differ case by case, and direct confirmation with BOI is required. The relationship between existing BOI privileges already held by a business and the new measures also needs working through per establishment.

The practical implication is this. If the project is large enough that factoring in BOI privileges would change the investment decision, start the confirmation with BOI before issuing the RFP. Begin checking the scheme only once quotations are in and you may miss the application deadline. Whether the machinery import duty exemption applies also depends on how gateways and sensors are classified, so decide who handles the import formalities for the hardware at the point you request quotations.

Implementation sequence – the first 120 days

Here is everything above rearranged into execution order. The goal is to reach, within 120 days, a state where quotations can be compared and where measured PoC figures exist. Do not set completion of the production rollout as the 120-day goal.

Two things are deliberately placed late in this sequence.

The first is requesting quotations. Waiting until day 51 is necessary because unless signal source classification and measurement requirements are settled in the preceding 50 days, the quotations that come back cannot be compared. Issue the RFP first and agree to work out the details later, and the money moves when you do work them out, at which point the comparison has lost its meaning.

The second is the production rollout decision. Do not fix the rollout total until day 120, when measured PoC figures exist. Per-machine effort can only be measured in a PoC, and every number before that is an estimate.

One thing is deliberately placed early. Agreeing the value of one hour of downtime with finance by day 35. That number is determined independently of factory IoT cost and does not need to wait for vendor selection. Without it, the payback calculation cannot be performed and the capital request will not clear internal approval. Many projects start hunting for this number only after the quotations are in, and stall there for weeks.

Frequently asked questions

What is the minimum budget to start factory IoT?

Starting from a budget figure will always mislead you. The smallest workable configuration is to narrow the target to three machines and run one thread through Layers 1 to 5. Under the assumptions used in this article, building the Scenario B configuration on three machines would not cost a tenth of the 30-machine figure, because the Layer 3 collection platform and network do not change much with machine count. That is a structural point from this article’s build-up, not a market rate. The real floor depends on how much of your existing network can be reused. The question “how little can we start with” is better replaced with “how much of Layer 3 are we building new”, and then it has an answer.

Can we compare multiple vendors on price per machine?

No. Layer 1 signal sources differ machine by machine and Layer 2 installation conditions differ site by site, and a per-machine unit price averages those differences away. What belongs in the denominator is the set of signal point count, sampling interval and tolerance for missing data. Align those three first, then line up the money for each of the five layers separately. If any layer is blank on a quotation, somebody will be doing that work after the contract is signed.

How do we avoid the project ending at a PoC?

Limit the PoC by machine count, never by layer. Three machines is enough, provided you run from Layer 1 through Layer 5 and include downtime reason entry. A PoC that skips Layer 4 and builds only Layers 3 and 5 gets a warm reception because the dashboard appears quickly, but it produces no basis for the rollout quotation. There are three criteria. Did downtime reason entry run as routine daily work for at least a month? Are the equipment master and tag design in a form usable in production as-is? Can per-machine effort be estimated? Answer all three and the PoC has succeeded.

How do we prevent a second round of cost?

Include Layer 4 in the initial quotation. Write tag design, equipment master, units, shift calendar and downtime reason codes into the RFP as mandatory items. Omit Layer 4 and the initial cost falls, but you end up with a system that displays the colour of a status lamp, and within about six months it becomes clear it cannot be used for improvement work. Retrofitting meaning from that point takes more effort than designing it on a clean sheet. This article assumed an additional 300,000 baht in year 2 for Scenario A. That is our own build-up rather than a market rate, but the structural point, that the second round can reach a third of the initial cost, holds regardless of the site.

Can BOI privileges be used for a factory IoT investment?

The Thailand Board of Investment announced new investment incentives on 15 January 2026, and the framework for promoting technology upgrades sets out an additional three years of corporate income tax exemption and an import duty exemption on machinery. Eligible technologies explicitly include predictive maintenance systems and IoT sensor platforms. However, eligibility conditions, the scope of qualifying machinery and application deadlines differ case by case, so direct confirmation with BOI is required. This article does not state specific deduction rates or monetary effects. In practice, if the project is large enough that the presence or absence of privileges would change the investment decision, begin the confirmation before issuing the RFP.

What value should we assign to one hour of downtime?

Use your own number. This article presents 50 baht per hour, derived from the Bangkok minimum wage of 400 baht per day divided by eight hours, as a floor. That figure is nothing more than the wage of one operator standing idle and cannot serve as a basis for return. The real loss is dominated by the gross margin on the product that machine could have made, in other words foregone profit. Derive your own figure from standard cost, from the machine hour rate, or from the current order backlog. You do not need a factory IoT quotation to settle this number, and it can be agreed with finance at the very start of the project.

Summary

Here are the points this article has set out on factory IoT cost.

• Cost breaks into five layers. Layer 1 signal source, Layer 2 connection and installation, Layer 3 collection platform, Layer 4 meaning, Layer 5 the user-facing side. Before comparing quotations, confirm which layers each vendor has included.
• Layer 1 sets the order of magnitude. Ethernet on the PLC, a PLC that is serial or undocumented, dry contacts only, or nothing at all. Counting your target machines against those four categories is the first piece of work.
• Layer 2 creates the spread. Number of shutdown windows, area classification, handling of existing warranty. Any quotation produced without a site survey necessarily contains standard assumptions.
• Layer 4 is what the second round of cost really is. Skip tag design, equipment master and downtime reason codes and the system still runs at first, but it cannot be used for improvement.
• Price per machine cannot be used for comparison. The denominator is the set of signal point count, sampling interval and tolerance for missing data.
• In the build-up for 30 machines at a Thai site, three-year total cost of ownership ranks A, B, C, but ranked on cost per usable month the order becomes B, C, A. The configuration that started cheapest ends up the most expensive. These figures are this article’s own build-up, not market rates.
• Put the return on reduced downtime, and set the value of one hour of downtime yourself. The 50 baht per hour derived from the Bangkok minimum wage of 400 baht per day is a floor, not a basis for return.
• BOI announced new investment incentives on 15 January 2026, with an additional three years of corporate income tax exemption and an import duty exemption on machinery under the technology upgrade framework. IoT sensor platforms are among the eligible technologies, but conditions and deadlines require confirmation with BOI.

Factory IoT quotations look like they differ by an order of magnitude not because vendor pricing is opaque, but because the buyer requested quotations without defining the signal sources on the machines and the measurement requirements. Settle those two and quotations become comparable.

TOMAS TECH builds equipment data collection platforms for Japanese-owned plants in Thailand. We can help with taking stock of signal sources before anyone opens a control panel, organising measurement requirements, and reading quotations decomposed into the five layers, all against the conditions at your site. It is fine to come to us before you have approached a single vendor, and before you have decided how many machines are in scope. If you would like to start by finding out how much signal your existing equipment can actually give you, get in touch through our contact page.

References

• GXO, cost benchmarks for developing an IoT data collection and visualisation platform, 2026 edition
• Three-up Technology, factory IoT implementation cost, duration and approach
• Yachiyo Solutions, factory IoT implementation methods, steps, sensor selection, cost and effect measurement
• Conexio, what factory IoT is, its benefits, cost and procedure
• Next Vision, why factory DX stalls at PoC – the structural reasons a successful trial never reaches production
• Alvarez and Marsal, Thailand’s Renewed BOI Incentives – A Strategic Window for Growth, Expansion and Investment 2026-2027
• Pertama Partners, BOI Manufacturing and Industry 4.0 Thailand 2026
• JETRO, Bangkok minimum wage raised to 400 baht per day